CN110365518A - Virtual machine bandwidth allocation methods of the OpenStack based on application service - Google Patents
Virtual machine bandwidth allocation methods of the OpenStack based on application service Download PDFInfo
- Publication number
- CN110365518A CN110365518A CN201910487936.4A CN201910487936A CN110365518A CN 110365518 A CN110365518 A CN 110365518A CN 201910487936 A CN201910487936 A CN 201910487936A CN 110365518 A CN110365518 A CN 110365518A
- Authority
- CN
- China
- Prior art keywords
- model
- virtual machine
- bandwidth
- sequence
- application
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 58
- 230000007812 deficiency Effects 0.000 claims abstract description 5
- 238000005311 autocorrelation function Methods 0.000 claims description 83
- 241001123248 Arma Species 0.000 claims description 35
- 238000004364 calculation method Methods 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 19
- 206010019133 Hangover Diseases 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims description 6
- 238000007476 Maximum Likelihood Methods 0.000 claims description 5
- 238000000780 augmented Dickey–Fuller test Methods 0.000 claims description 5
- 239000000203 mixture Substances 0.000 claims description 5
- 238000004064 recycling Methods 0.000 claims description 4
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 238000009825 accumulation Methods 0.000 claims description 3
- 238000013459 approach Methods 0.000 claims description 3
- 238000011835 investigation Methods 0.000 claims description 3
- 238000012417 linear regression Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims 1
- 230000003068 static effect Effects 0.000 abstract description 2
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 abstract 1
- 239000004615 ingredient Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000032258 transport Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000000977 Dickey–Fuller test Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/08—Configuration management of networks or network elements
- H04L41/0896—Bandwidth or capacity management, i.e. automatically increasing or decreasing capacities
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/4557—Distribution of virtual machine instances; Migration and load balancing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
- G06F9/45533—Hypervisors; Virtual machine monitors
- G06F9/45558—Hypervisor-specific management and integration aspects
- G06F2009/45595—Network integration; Enabling network access in virtual machine instances
Abstract
A kind of virtual machine bandwidth allocation methods the invention discloses OpenStack based on application service, this method comprises the following steps: step 1, the network QoS index of specified OpenStack cloud platform application;The determination of step 2, ARIMA prediction model;Step 3 calculates the effective bandwidth based on application QoS according to application stream statistics characteristic;Step 4, the network load state that a virtual machine lower period is predicted according to effective bandwidth;Step 5, by the network load state of virtual machine according to different allocation strategy bandwidth allocations.Virtual machine Bandwidth Dynamic Allocation method provided by the invention, overcome the low disadvantage of static bandwidth allocation method resource utilization, compensating for traditional bandwidth distribution method not can guarantee the deficiency of application network QoS, so that OpenStack cloud platform bandwidth resources are rationally and efficiently used, bandwidth resources utilization rate is improved, while ensure that the QoS of virtual machine application.
Description
Technical field
The present invention relates to field of cloud calculation, and in particular to a kind of virtual machine bandwidth point of the OpenStack based on application service
Method of completing the square.
Background technique
Cloud computing is the delivery mode and business model of novel computing resource, Internet resources and storage resource, by dynamic
The telescopic virtualization resource of state improves service.The open source cloud platform item that OpenStack remains unchanged surging as current liveness
Mesh relies on its efficient community development and flexible deployment mode, obtains the extensive concern of business circles and academia.
Internet resources are important one of the resources of cloud data center, and the Internet resources of cloud data center access are limited, such as
What reasonable distribution Internet resources, is cloud data center problem to be solved.OpenStack is deployed in physical server at present
On, the bandwidth of data center's access based on OpenStack is limited, and the unreasonable distribution of resource will lead on physical node
Virtual machine access outer net or when other virtual machines contention access resource, the base application service that cloud platform provides can not be by
It need to be rationally using the Internet resources of distribution, so as to cause the waste of resource.Therefore, how to dynamically distribute access OpenStack's
One of the problem of bandwidth resources are cloud platform urgent need to resolve.
Summary of the invention
In view of the above problems, the present invention provides a kind of just virtual machine Bandwidth Dynamic Allocation method based on application service, it can
Effectively to solve OpenStack cloud platform virtual machine to the race problem of bandwidth resources, guaranteeing the same of the network QoS of application
Shi Tigao resource utilization.
A kind of virtual machine bandwidth allocation methods of the OpenStack based on application service, the described method comprises the following steps:
S1, network QoS index (Quality of Service, the Service Quality for formulating the application of OpenStack cloud platform
Amount);
S2, ARIMA prediction model is determined;
S3, the effective bandwidth based on application QoS is calculated according to application stream statistics characteristic;
S4, the network load state that a virtual machine lower period is predicted according to effective bandwidth;
S5, the network load state according to virtual machine give virtual machine bandwidth allocation.
Preferably, the network QoS index of OpenStack cloud platform application is formulated in step S1, comprising the following steps:
S1-1, the business network QoS demand index for investigating different company.Business is divided into voice service, video traffic, number
According to business, different applications has different network QoS index, including time delay, shake, packet loss etc. under each business.It is similar
Finding is as shown in table 1, table 2, table 3, table 4.
The QoS of survice demand parameter that 1 Huawei Tech Co., Ltd of table proposes
The QoS of survice demand parameter that 2 Wuhan Yangtze Optical Technology Co., Ltd of table proposes
The QoS of survice demand characteristic that 3 Fiberhome Telecommunication Tech Co., Ltd. of table proposes
The QoS of survice demand parameter that 4 Fiberhome Telecommunication Tech Co., Ltd. of table proposes
S1-2, an application is often increased newly, administrator refers to the index of S1-1 investigation, the network QoS of the application is specified to refer to
Mark, the basic element of each application are as shown in table 5 below.
5 OpenStack cloud platform application virtual machine of table describes table basic element
Preferably, ARIMA (the Autoregressive Integrated Moving Average model) model
Mainly it is made of autoregression model AR model, difference operator d and moving average model(MA model) MA model.
AR (Auto-regressive) model, that is, autoregression model, predicts this with the performance situation before same variable
The present or following performance situation of variable, it is assumed that AR model is p rank, is indicated with AR (p), for a time series X
(t), if meeting
Wherein, XtFor time series t moment observed value,For autoregression system to be estimated
Number, μtFor stochastic error;
MA (Moving Average) model, that is, moving average model(MA model) show that q rank is moved using the error between sequential digit values
Dynamic averaging model, is indicated, expression formula with MA (q) are as follows:
Xt=μt-θ1μt-1-θ2μt-2-…-θqμt-q (2)
Wherein, θ1, θ2..., θqFor rolling average coefficient to be estimated, μt, μt-1, μt-2..., μt-qIt is model in t
The error at moment, t-1 moment ..., t-q moment;
Arma modeling, that is, autoregressive moving-average model predicts that change in future becomes by characteristic that time series has
Gesture indicates that ARMA (p, q) model is the combination of AR (p) model and MA (q) model with ARMA (p, q), ARMA (p, q) model its
Expression formula are as follows:
AR model, MA model and arma modeling and for non-stationary series, are needed premised on stationary sequence
Integrate rolling average autoregression model, also referred to as ARIMA model with difference, ARIMA model mainly first to non-stationary series into
D Difference Calculation of row is converted to stationary sequence, then carries out model identification, is denoted as ARIMA (p, d, q);
Whether steady judge flow sequence by stationary test, if steadily, d=0, at this time ARIMA (p, d, q) be
ARIMA (p, 0, q), is equal to ARMA (p, q);If non-stationary, flow sequence is converted to steady sequence after d Difference Calculation
Column are converted to for ARMA (p, q) model after stationary sequence, and as p=0, ARMA (p, q) is degenerated to MA (q) model, works as q=0
When, ARMA (p, q) is degenerated to AR (p) model.
In the determination of step S2 ARIMA prediction model, the specific steps are as follows:
S2-1, the flow for acquiring virtual machine carry out ADF (Augmented Dickey-Fuller test) to flow sequence
Stationary test carries out Difference Calculation and is converted to stationary sequence if sequence non-stationary;Difference Calculation process is as follows:
Non-stationary series XtDifference is converted into stationary sequence, and difference number d is also referred to as homogeneous rank, usesIndicate that difference is calculated
Son, first-order difference following formula subrepresentation:
Similarly, second differnce following formula subrepresentation:
It is general at most to be turned to stationary sequence by second order difference.
S2-2, the range that p, q are determined by auto-correlation function ACF and partial autocorrelation function PACF;
Which specifically, being converted to stationary sequence by stationary test flow sequence by S2-1, to judge specific using
Model, need to judge whether p, q are equal to 0, and determine the range of p, q.Auto-correlation function ACF and partial autocorrelation function PACF
It can determine p, q value range.The corresponding ACF and PACF feature of each concrete model is as shown in table 6 below.Truncation refers to flow sequence
The ACF or PACF of column are 0 property after certain rank, tend to 0 quickly after being greater than some constant k as the truncation of k rank.Hangover is
ACF or PACF is not 0 property after certain rank, i.e., has non-zero value always, will not be just permanent after k is greater than some constant
Equal to zero.
Concrete model corresponding ACF and PACF under 6 ARIMA model of table
(1) calculation formula of ACF and PACF is such as are as follows:
Time series XtLag the sample autocorrelation function (ACF) of k rank are as follows:
T indicates time point, and T indicates time series final time point,Indicate the time series in time point k+1 to T
Average value;
WithJ-th of regression coefficient in k rank autoregression formula, j=1~k are indicated, then k rank autoregression model indicates are as follows:
It is the last one regression coefficient, the function of lag period k are as follows:
It is denoted by partial autocorrelation function, i.e., by the following formulaComposition.
WithYule-Walker equation is expressed, sample partial autocorrelation function (PACF) under lag k rank can be derived from
Are as follows:
(2) ACF and PACF of MA (q) model
Auto-correlation function is used to determine the order q of moving average model(MA model), defines auto-covariance function rkIndicate lag k's
Stochastic variable XtAnd Xt-kBetween degree of correlation, then two variable Xs of k rankt、Xt-kBetween covariance be
rk=cov (Xt, Xt-k)=E [(Xt-μ)(Xt-k-μ)] (13)
μ indicates the average value of the sequence at t-k to t time point, i.e.,
σ2It is the variance of white noise sequence, then auto-correlation function ρkAbbreviation are as follows:
r0Covariance r when k=0 in expression (14)kValue, to obtain
It is therefore seen that the ACF of MA (q) is 0 at k > q, that is, ACF is q step truncation.And its PACF is no matter k takes
Much, the calculated value of PACF is related with the auto-correlation function of 1 to q rank lag.
(3) ACF and PACF of AR (p) model
For AR (p) model expression
Its ACF is
Wherein
Then
It can be seen that the ACF of AR (p) no matter k take it is much, to 1 to p rank lag ACF it is related, be not 0, to drag
Tailer sequence.
For PACF, can be calculated according to following:
For AR (1) process, as k=1,As k > 1,Namely occurs peak value then truncation in k=1.
For AR (2) process, partial autocorrelation function is made of formula (9) and formula (10), it can be seen that in k≤2,As k > 2,That is the truncation after k=2.
Therefore, it can release for AR (p) process, in k≤p,As k > p, the PACF of AR (p) is
0, that is, PACF is with the truncated sequence of p rank truncation.
(4) ACF and PACF of ARMA (p, q) model
The ACF of ARMA (p, q) model can regard the mixing of the ACF of the ACF and AR (p) of MA (q) as.If p=0, it is
Truncated sequence, if q=0, it is hangover sequence, if p and q are not 0, it is hangover sequence.
The PACF of ARMA (p, q) model can equally regard the mixing of the PACF of the PACF and AR (p) of MA (q) as.If p
=0, it is hangover sequence, if q=0, it is truncated sequence, if p and q are not 0, it is hangover sequence.
S2-3, by the range of S2-2 step available p and q, different p, q values is combined in the range, generation
Enter into AIC criterion function and calculated, obtain the value of p, q when AIC minimum, so that it is determined that optimal ARIMA (p, d, q) mould
Type;AIC criterion function is as follows:
AIC=-2ln (L)+2k (21)
Wherein L is the maximum likelihood function value of ARIMA model, and k is the estimative number of parameters of ARIMA model.Such as
S2-1 step passes through d1(d1>=0) flow sequence is converted to stationary sequence by secondary Difference Calculation, is chosen within the scope of p, q and is arrived some
Combine (p1, q1), then ARIMA model is ARIMA (p1, d1, q1), calculate corresponding maximum likelihood function value L1, k is in formula (3)
The number of parameter to be estimated.
Preferably, step S3 calculates the effective bandwidth based on application QoS, including following step according to application stream statistics characteristic
It is rapid:
S3-1, recalibration domain method calculate the Hurst index value H of flow sequence;
Specifically, recalibration domain method is also referred to as R/S analytic approach, the basic principle of gauge index value H is as follows: (1) giving
The time series X (t) that one length is N, dividing equally entire sequence with length n is A adjacent subintervals, then has A × n=N;
Any subinterval is expressed as Ia, a=1,2 ..., A, in IaIn element representation be N (k, m), k=1,2 ..., n;m
=1,2 ... .., A, IaMean value be
(2)IaFor the accumulation intercept X of mean valueK, aIt is defined as follows:
NI, aIt indicates in IaIn element, i=1,2 ..., n
(3)IaIt is very poor
(4) subinterval IaStandard deviation
(5) eachBy correspondingIt is standardized, obtains sequence (R/S) n:
(7) n is since 3, and repeats step (1)-step (5) and obtain sequence [R/S] until n=Nn, n=3 ...,
N;
(7) with Log (n) for explanatory variable, Log (R/S) is that explained variable carries out linear regression:
Log (R/S)=Loge+HLogn (26)
Wherein, e is the normal number randomly selected, and slope H is Hurst index value.
S3-2, the effective bandwidth flowed according to the statistical property of application stream, calculating application;
Specifically, Y (t) is application traffic sequence according to the statistical property of network flow, it can see that ingredient shape Blang transports
Dynamic process, then have:
Wherein, ZtIt is normal state fractal Brown motion, is also zero-mean, 0 < H < 1 and t of self-similarity nature parameter2HFor variance
Standard fractal Brown motion, y is flow serial mean, and m is coefficient of variation, and t indicates time variable, and H is Hurst index
Value.
There is formula (27) to derive that queue buffer queue length is greater than the probability of b:
Wherein, C is effective bandwidth, k (H)=HH(1-H)1-H.P (X (t) > x) is equal to packet loss ε, and b indicates that queue is slow
Qu great little is rushed, B (t) indicates that data reach the data length of buffer area with time change, substituted into after formula (28) and obtained with ε
The effective bandwidth C flowed to application:
Tolerant of delay and packet loss are considered in effective bandwidth calculating, overcome effective bandwidth originally based on Loss Rate
Calculation method cannot be guaranteed that the deficiency of time delay, binding cache time delay maximum value=queue buffer storage length ÷ queue output rate obtain
To formula (15)
B=y × τ (30)
Wherein τ indicates the tolerant of delay of application stream, is substituted into formula (29) and obtains meeting the effective of network QoS demand
Bandwidth are as follows:
Wherein k=HH(1-H)1-H。
Preferably, network load state described in step S4 is measured with network utilization μ, and network utilization μ is void
The effective bandwidth of quasi- machine application stream and administrator are the ratio of the bandwidth of virtual machine distribution;
Divide the network load state of virtual machine to low network load, high network load and netful network according to network utilization
Load, wherein if low network load indicates that following a period of time distributes current bandwidth, virtual machine network utilization rate μ will be small
In 70%, it is determined as the state few in a concurrency, load is small, therefore transmitted data amount is small, Internet resources are opposite
Free time needs to recycle at this time the bandwidth of 1- (μ+10%);If high network load indicates that following a period of time distributes current band
Width, virtual machine network utilization rate μ will be greater than or equal to 70% but be less than or equal to 85%, be determined as virtual machine be in network
Demand requires relatively high state, but can make full use of bandwidth resources at this time, and user meets the network of setting using application
Demand, the bandwidth of application service virtual machine is without changing at this time;If full network load indicates following a period of time virtual machine point
With current bandwidth, virtual machine network utilization rate μ will be greater than 85%, determine the virtual machine shape relatively high to network resource requirements
State, and the bandwidth of application service virtual machine setting can not meet network demand, the bandwidth of application service virtual machine needs at this time
It increases, initial increase is 1.1 times of original bandwidth allocation.
It preferably, is by current period bandwidth allocation E in the method for salary distribution of step S51, effective bandwidth C2According to formulaFuture period virtual machine network utilization rate μ is calculated, indicates the bandwidth that a virtual machine lower period should distribute with E, according to
μ value range judges the future period network state of virtual machine, with β and γ to dividing using service virtual machine, obtain β=
0.7, γ=0.85
If 0 < μ < β, virtual machine carries out bandwidth recycling, recycles the bandwidth of 1- (μ+10%), then E=E1×(μ+
10%) (32)
If β≤μ≤γ, virtual machine bandwidth is remained unchanged, then
E=E1(33)
If μ > γ, virtual machine bandwidth expands, then
E=E1× (1+10%) (34).
The present invention is had the following advantages and beneficial effects: compared with the prior art
A kind of virtual machine bandwidth allocation methods the present invention provides OpenStack based on application service, can effectively solve
Certainly OpenStack cloud platform virtual machine improves money to the race problem of bandwidth resources while guaranteeing the network QoS of application
Source utilization rate.The characteristic of network application stream, and the application QoS index for combining each major company to propose are analyzed in the present invention first
And the usage scenario of OpenStack cloud platform, table is described provided with the QoS index for the application of OpenStack cloud platform, so
The calculation of connected applications flow effective bandwidth afterwards joined tolerant of delay, tolerate that packet loss improves, finally chooses
The Time series analysis method ARIMA model of comparative maturity predicts flow, and predicts virtual machine according to effective bandwidth
The network load state in next period uses different allocation strategies by different network loads, thus to the band of virtual machine
Width, which is realized, to be dynamically distributed.
Detailed description of the invention
Fig. 1 is virtual machine Bandwidth Dynamic Allocation strategic process figure of the embodiment of the present invention based on application service.
Specific embodiment
Below with reference to embodiment, the invention will be further described, but does not constitute any limitation of the invention, appoints
Where the modification for the limited times that scope of the invention as claimed is made, still in scope of the presently claimed invention.
A kind of virtual machine bandwidth allocation methods of the OpenStack of the present embodiment based on application service, the method includes
Following steps:
Step 1, the network QoS index for formulating the application of OpenStack cloud platform, network QoS refer to guarantee certain networks
For demand to meet the ability of the Service Level Agreement between application provider and end user (SLA), network demand includes band
Wide, delay, shake and reliability.
Formulation process is specific as follows:
S1-1, the business network QoS demand index for investigating different company.Business is divided into voice service, video traffic, number
According to business, different applications has different network QoS index, including time delay, shake, packet loss etc. under each business.It is similar
Finding is as shown in table 1, table 2, table 3, table 4.
The QoS of survice demand parameter that 1 Huawei Tech Co., Ltd of table proposes
The QoS of survice demand parameter that 2 Wuhan Yangtze Optical Technology Co., Ltd of table proposes
The QoS of survice demand characteristic that 3 Fiberhome Telecommunication Tech Co., Ltd. of table proposes
The QoS of survice demand parameter that 4 Fiberhome Telecommunication Tech Co., Ltd. of table proposes
S1-2, an application is often increased newly, administrator refers to the index of S1-1 investigation, the network QoS of the application is specified to refer to
Mark, the basic element of each application are as shown in table 5 below.
5 OpenStack cloud platform application virtual machine of table describes table basic element
Step 2, the lower period flow sequence of ARIMA model prediction, the specific steps are as follows:
The data detection of S2-1, flow sequence.The flow for acquiring virtual machine carries out ADF (Augmented to flow sequence
Dickey-Fuller test) stationary test, if sequence non-stationary, progress Difference Calculation is converted to stationary sequence;
Specifically, ARIMA (Autoregressive Integrated Moving Average model) model is main
It is made of autoregression model AR model, difference operator d, moving average model(MA model) MA model.
AR (Auto-regressive) model, that is, autoregression model, predicts this with the performance situation before same variable
The present or following performance situation of variable, this prediction technique is only related with oneself, and unrelated with its dependent variable, referred to as
Autoregression.It is assumed that AR model is p rank, then for a time series X (t), if meeting
Wherein, XtFor time series t moment observed value,For autoregression system to be estimated
Number, μtFor stochastic error.
MA (Moving Average) model, that is, moving average model(MA model), the model are obtained using the error between sequential digit values
Q rank moving average model(MA model) out is indicated, expression formula with MA (q) are as follows:
Xt=μ t- θ1μt-1-θ2μt-2-…-θqμt-q (2)
Wherein, θ1, θ2..., θqFor rolling average coefficient to be estimated, μt, μt-1, μt-2..., μt-qIt is model in t
The error at moment, t-1 moment ..., t-q moment.
Arma modeling, that is, autoregressive moving-average model predicts that change in future becomes by characteristic that time series has
Gesture, ARMA (p, q) model are the combinations of AR (p) model and MA (q) model, and ARMA (p, q) is expressed as
AR model, MA model, arma modeling are all and for non-stationary series, to need to use premised on stationary sequence
Difference integrates rolling average autoregression model, also referred to as ARIMA model, mainly first carries out d difference to non-stationary series
Calculating is converted to stationary sequence, then carries out model identification.Usually it is denoted as ARIMA (p, d, q).
It to sum up analyzes, i.e. ARIMA (p, d, q) model is universal model.Whether flow sequence is judged by stationary test
Steadily, if steadily, d=0, ARIMA (p, d, q) is ARIMA (p, 0, q) at this time, it is equal to ARMA (p, q).If non-stationary,
Flow sequence is converted to stationary sequence after d Difference Calculation, and being converted to after stationary sequence is exactly ARMA (p, q) model,
As p=0, ARMA (p, q) is degenerated to MA (q) model, and as q=0, ARMA (p, q) is degenerated to AR (p) model.
Difference Calculation process is as follows:
Non-stationary series XtDifference is converted into stationary sequence, and difference number d is also referred to as homogeneous rank.WithIndicate that difference is calculated
Son, then first-order difference following formula subrepresentation:
Similarly, second differnce following formula subrepresentation:
It is general at most to be turned to stationary sequence by second order difference
S2-2, the range that p, q are determined by auto-correlation function ACF and partial autocorrelation function PACF;
Which specifically, being converted to stationary sequence by stationary test flow sequence by S2-1, to judge specific using
Model, need to judge whether p, q are equal to 0, and determine the range of p, q.Auto-correlation function ACF and partial autocorrelation function PACF
It can determine p, q value range.The corresponding ACF and PACF feature of each concrete model is as shown in table 6 below.Truncation refers to flow sequence
The ACF or PACF of column are 0 property after certain rank, tend to 0 quickly after being greater than some constant k as the truncation of k rank.Hangover is
ACF or PACF is not 0 property after certain rank, i.e., has non-zero value always, will not be just permanent after k is greater than some constant
Equal to zero.
Concrete model corresponding ACF and PACF under 6 ARIMA model of table
The derivation process of table 6 is as follows:
(1) calculation formula of ACF and PACF
Time series XtLag the sample autocorrelation function (ACF) of k rank are as follows:
T indicates time point, and T indicates time series final time point,Indicate the time series in time point k+1 to T
Average value;
WithJ-th of regression coefficient in k rank autoregression formula, j=1~k are indicated, then k rank autoregression model indicates are as follows:
It is the last one regression coefficient, the function of lag period k are as follows:
It is denoted by partial autocorrelation function, i.e., by the following formulaComposition.
WithYule-Walker equation is expressed, sample partial autocorrelation function (PACF) under lag k rank can be derived from
Are as follows:
(2) ACF and PACF of MA (q) model
Auto-correlation function is used to determine the order q of moving average model(MA model), defines auto-covariance function rkIndicate lag k's
Stochastic variable XtAnd Xt-kBetween degree of correlation, then being separated by two variable Xs of k rankt、Xt-kBetween covariance be
rk=cov (Xt, Xt-k)=E [(Xt-μ)(Xt-k-μ)] (13)
μ indicates the average value of the sequence at t-k to t time point.
I.e.
σ2It is the variance of white noise sequence.
So auto-correlation function ρkAbbreviation are as follows:
r0Covariance r when k=0 in expression (14)kValue, to obtain
It is therefore seen that the ACF of MA (q) is 0 at k > q, that is, ACF is q step truncation.And its PACF is no matter k takes
Much, the calculated value of PACF is related with the auto-correlation function of 1 to q rank lag.
(3) ACF and PACF of AR (p) model
For AR (p) model expression
Its ACF is
Wherein
Then
It can be seen that the ACF of AR (p) no matter k take it is much, to 1 to p rank lag ACF it is related, be not 0, to drag
Tailer sequence.
For PACF, can be calculated according to following:
For AR (1) process, as k=1,As k > 1,Namely occurs peak value then truncation in k=1.
For AR (2) process, partial autocorrelation function is made of formula (9) and formula (10), it can be seen that in k≤2,As k > 2,That is the truncation after k=2.
Therefore, it can release for AR (p) process, in k≤p,As k > p, the PACF of AR (p) is
0, that is, PACF is with the truncated sequence of p rank truncation.
(4) ACF and PACF of ARMA (p, q) model
The ACF of ARMA (p, q) model can regard the mixing of the ACF of the ACF and AR (p) of MA (q) as.If p=0, it is
Truncated sequence, if q=0, it is hangover sequence, if p and q are not 0, it is hangover sequence.
The PACF of ARMA (p, q) model can equally regard the mixing of the PACF of the PACF and AR (p) of MA (q) as.If p
=0, it is hangover sequence, if q=0, it is truncated sequence, if p and q are not 0, it is hangover sequence.
S2-3, by the range of S2-2 step available p and q, different p, q values is combined in the range, generation
Enter into AIC criterion function and calculated, obtain the value of p, q when AIC minimum, so that it is determined that optimal ARIMA (p, d, q) mould
Type;AIC criterion function is as follows:
AIC=-2ln (L)+2k (21)
Wherein L is the maximum likelihood function value of ARIMA model, and k is the estimative number of parameters of ARIMA model.Such as
S2-1 step passes through d1(d1>=0) flow sequence is converted to stationary sequence by secondary Difference Calculation, is chosen within the scope of p, q and is arrived some
Combine (p1, q1), then ARIMA model is ARIMA (p1, d1, q1), calculate corresponding maximum likelihood function value L1, k is in formula (3)
The number of parameter to be estimated.
Step 3 calculates the effective bandwidth based on application QoS according to application stream statistics characteristic, step specific as follows:
S3-1, recalibration domain method calculate the Hurst index value H of flow sequence;
Specifically, recalibration domain method is also referred to as R/S analytic approach, the basic principle of gauge index value H is as follows:
(1) the time series X (t) that a length is N is given, dividing equally entire sequence with length n is A adjacent sub-districts
Between, then there is A × n=N.Any subinterval is expressed as Ia, a=1,2 ..., A.In IaIn element representation be N (k, m), k
=1,2 ..., n;M=1,2 ..., A.IaMean value be
(2)IaFor the accumulation intercept X of mean valueK, aIt is defined as follows:
(3)IaIt is very poor
(4) subinterval IaStandard deviation
(5) eachBy correspondingIt is standardized, obtains sequence (R/S)n:
(6) n is since 3, and repeats 1-5 step, until n=N, obtains sequence [R/S]n, n=3 ..., N.
(7) with Log (n) for explanatory variable, Log (R/S) is explained variable progress linear regression: Log (R/S)=
Loge+H·Logn(26)
Wherein, e is the normal number randomly selected, and slope H is Hurst index value.
S3-2, the effective bandwidth flowed according to the statistical property of application stream, calculating application;
Specifically, Y (t) is application traffic sequence according to the statistical property of network flow, it can see that ingredient shape Blang transports
Dynamic process, then have:
Wherein, ZtIt is normal state fractal Brown motion, is also zero-mean, 0 < H < 1 and t of self-similarity nature parameter2HFor variance
Standard fractal Brown motion, y is flow serial mean, and m is coefficient of variation, and t indicates time variable, and H is Hurst index
Value.
There is formula (27) to derive that queue buffer queue length is greater than the probability of b:
Wherein, C is effective bandwidth, k (H)=HH(1-H)1-H.P (X (t) > x) is equal to packet loss ε, and b indicates that queue is slow
Qu great little is rushed, B (t) indicates that data reach the data length of buffer area with time change, substituted into after formula (28) and obtained with ε
The effective bandwidth C flowed to application:
Tolerant of delay and packet loss are considered in effective bandwidth calculating, overcome effective bandwidth originally based on Loss Rate
Calculation method cannot be guaranteed that the deficiency of time delay, binding cache time delay maximum value=queue buffer storage length ÷ queue output rate obtain
To formula (30)
B=y × τ (30)
Wherein τ indicates the tolerant of delay of application stream, is substituted into formula (29) and obtains meeting the effective of network QoS demand
Bandwidth are as follows:
Wherein k=HH(1-H)1-H。
Step 4, the network load state that a virtual machine lower period is predicted according to effective bandwidth;The network load shape
State network load state is measured with network utilization μ, and network utilization μ is the effective bandwidth and pipe of virtual machine application stream
Reason person is the ratio of the bandwidth of virtual machine distribution;
Divide the network load state of virtual machine to low network load, high network load and netful network according to network utilization
Load, wherein if low network load indicates that following a period of time distributes current bandwidth, virtual machine network utilization rate μ will be small
In 70%, it is determined as the state few in a concurrency, load is small, therefore transmitted data amount is small, Internet resources are opposite
Free time needs to recycle at this time the bandwidth of 1- (μ+10%);If high network load indicates that following a period of time distributes current band
Width, virtual machine network utilization rate μ will be greater than or equal to 70% but be less than or equal to 85%, be determined as virtual machine be in network
Demand requires relatively high state, but can make full use of bandwidth resources at this time, and user meets the network of setting using application
Demand, the bandwidth of application service virtual machine is without changing at this time;If full network load indicates following a period of time virtual machine point
With current bandwidth, virtual machine network utilization rate μ will be greater than 85%, determine the virtual machine shape relatively high to network resource requirements
State, and the bandwidth of application service virtual machine setting can not meet network demand, the bandwidth of application service virtual machine needs at this time
It increases, initial increase is 1.1 times of original bandwidth allocation.
Step 5, by the network load state of virtual machine according to different allocation strategy bandwidth allocations.The method of salary distribution is specific
Are as follows:
By current period bandwidth allocation E1, effective bandwidth C2According to formulaCalculate future period virtual machine network
Utilization rate μ indicates the bandwidth that a virtual machine lower period should distribute with E, the future period of virtual machine is judged according to μ value range
Network state is divided to using service virtual machine with β and γ, obtains β=0.7, γ=0.85
If 0 < μ < β, virtual machine carries out bandwidth recycling, recycles the bandwidth of 1- (μ+10%), then
E=E1× (μ+10%) (32)
If β≤μ≤γ, virtual machine bandwidth is remained unchanged, then
E=E1(33)
If μ > γ, virtual machine bandwidth expands, then
E=E1× (1+10%) (34)
Virtual machine Bandwidth Dynamic Allocation method provided by the invention overcomes static bandwidth allocation method resource utilization
Low disadvantage, compensating for traditional bandwidth distribution method not can guarantee the deficiency of application network QoS, so that OpenStack cloud platform
Bandwidth resources are rationally and efficiently used, and bandwidth resources utilization rate is improved, while ensure that the QoS of virtual machine application.
In conjunction with above step, the virtual machine Bandwidth Dynamic Allocation strategic process figure based on application service is as shown in Figure 1, tool
Body the following steps are included:
Step 1 distributes initial bandwidth for virtual machine, and the initial bandwidth of every application service virtual machine is specified by administrator
(initial bandwidth as shown in table 5), and monitor virtual machine Microsoft Loopback Adapter changes in flow rate;
Step 2, using the timer in program every 30 minutes bandwidth calculations of execution and distribution;
Step 3, the time for reaching timer setting obtain the historical traffic of current period virtual machine are as follows:
F={ f1, f2, f3..., ft-1, ft}
Wherein t indicates time point, ftIt indicates in t time point collected data on flows.
Step 4, the flow sequence that an ARIMA model prediction lower period is utilized by the flow sequence that step 3 obtains are as follows:
P={ p1, p2, p3..., pt-1, pt}
Wherein t indicates time point, ptIndicate prediction in the data on flows of t time point virtual machine
Step 5, the flow sequence obtained by step 4 calculate the effective bandwidth C in next period2Are as follows:
Step 6, by current period bandwidth allocation E1, effective bandwidth C2According to formulaIt is virtual to calculate future period
Machine network utilization μ judges the future period network state of virtual machine according to μ value range.
If 0 < μ < β, virtual machine carries out bandwidth recycling, recycles the bandwidth of 1- (μ+10%), then
E=E1× (μ+10%) (36)
The wherein bandwidth that an E expression virtual machine lower period should distribute.
If β < μ < γ, virtual machine bandwidth remain unchanged, then
E=E1(37)
If μ > γ, virtual machine bandwidth expands, then
E=E1× (1+10%) (38).
In conclusion the above is only the preferred embodiment of the present invention, it should be pointed out that come for those skilled in the art
It says, without departing from the structure of the invention, several modifications and improvements can also be made, these all will not influence the present invention
The effect and practicability of implementation.
Claims (7)
1. a kind of virtual machine bandwidth allocation methods of OpenStack based on application service, which is characterized in that the method includes with
Lower step:
S1, the network QoS index for formulating the application of OpenStack cloud platform;
S2, ARIMA prediction model is determined;
S3, the effective bandwidth based on application QoS is calculated according to application stream statistics characteristic;
S4, the network load state that a virtual machine lower period is predicted according to effective bandwidth;
S5, the network load state according to virtual machine give virtual machine bandwidth allocation.
2. virtual machine bandwidth allocation methods OpenStack base of the OpenStack according to claim 1 based on application service
In the virtual machine bandwidth allocation methods of application service, which is characterized in that step S1 formulates the network of OpenStack cloud platform application
QoS index, comprising the following steps:
S1-1, the business network QoS demand index for investigating different company;Different applications has different networks under each business
QoS index;
S1-2, an application is often increased newly, administrator refers to the index of S1-1 investigation, specifies the network QoS index of the application, each
The basic element of application include: application ID, application name, application type, time delay, delay variation, packet loss, guarantee bandwidth and
Initial bandwidth.
3. virtual machine bandwidth allocation methods of the OpenStack according to claim 1 based on application service, feature exist
In ARIMA (the Autoregressive Integrated Moving Average model) model mainly by returning certainly
Return model AR model, difference operator d and moving average model(MA model) MA model composition;
AR (Auto-regressive) model, that is, autoregression model, predicts the variable with the performance situation before same variable
Present or following performance situation, it is assumed that AR model is p rank, is indicated with AR (p), for a time series X (t), if
Meet
Wherein, XtFor time series t moment observed value,For autoregressive coefficient to be estimated, μtFor
Stochastic error;
MA (Moving Average) model, that is, moving average model(MA model) show that q rank is mobile flat using the error between sequential digit values
Equal model is indicated, expression formula with MA (q) are as follows:
Xt=μt-θ1μt-1-θ2μt-2-…-θqμt-q (2)
Wherein, θ1,θ2,...,θqFor rolling average coefficient to be estimated, μt,μt-1,μt-2,...,μt-qIt is model in t moment,
The error at t-1 moment ..., t-q moment;
Arma modeling, that is, autoregressive moving-average model predicts future trends by characteristic that time series has, uses
ARMA (p, q) indicates that ARMA (p, q) model is the combination of AR (p) model and MA (q) model, its expression formula of ARMA (p, q) model
Are as follows:
AR model, MA model and arma modeling are all and for non-stationary series, to need to use difference premised on stationary sequence
Rolling average autoregression model, also referred to as ARIMA model are integrated, ARIMA model is mainly first poor to non-stationary series progress d times
Divide to calculate and be converted to stationary sequence, then carry out model identification, is denoted as ARIMA (p, d, q);
It is whether steady that flow sequence is judged by stationary test, if steadily, d=0, ARIMA (p, d, q) is ARIMA at this time
(p, 0, q), is equal to ARMA (p, q);If non-stationary, flow sequence is converted to stationary sequence after d Difference Calculation, conversion
To be ARMA (p, q) model after stationary sequence, as p=0, ARMA (p, q) is degenerated to MA (q) model, as q=0, ARMA
(p, q) is degenerated to AR (p) model.
4. virtual machine bandwidth allocation methods of the OpenStack according to claim 1 based on application service, feature exist
In the determination of step S2ARIMA prediction model, the specific steps are as follows:
It is steady to carry out ADF (Augmented Dickey-Fuller test) to flow sequence by S2-1, the flow for acquiring virtual machine
Property examine, if sequence non-stationary, carry out Difference Calculation be converted to stationary sequence;Difference Calculation process is as follows:
Non-stationary series XtDifference is converted into stationary sequence, and difference number d is also referred to as homogeneous rank, usesExpression difference operator, one
Order difference following formula subrepresentation:
Similarly, second differnce following formula subrepresentation:
S2-2, the range that p and q are determined by auto-correlation function ACF and partial autocorrelation function PACF;
(1) calculation formula of ACF and PACF are as follows:
Time series XtLag k rank sample autocorrelation function (ACF) be
T indicates time point, and T indicates time series final time point,It indicates to be averaged in the time series of time point k+1 to T
Value;
WithIndicate j-th of regression coefficient in k rank autoregression formula, j=1~k, then k rank autoregression model indicates are as follows:
It is the last one regression coefficient, the function of lag period k are as follows:
It is denoted by partial autocorrelation function, i.e., by the following formulaComposition;
WithYule-Walker equation is expressed, sample partial autocorrelation function (PACF) under lag k rank is derived from are as follows:
(2) ACF and PACF of MA (q) model
Auto-correlation function is used to determine the order q of moving average model(MA model), defines auto-covariance function rkIndicate the random of lag k rank
Variable XtAnd Xt-kBetween degree of correlation, then being separated by two variable Xs of k rankt、Xt-kBetween covariance be
rk=cov (Xt,Xt-k)=E [(Xt-μ)(Xt-k-μ)] (13)
μ indicates the average value of the sequence at t-k to t time point, i.e.,
σ2It is the variance of white noise sequence, auto-correlation function ρkAbbreviation are as follows:
r0Covariance r when k=0 in expression (14)kValue, to obtain
It is therefore seen that the ACF of MA (q) at k > q be 0, that is, ACF be q step truncation, and its PACF be no matter k take it is much,
The calculated value of PACF is related with the auto-correlation function of 1 to q rank lag;
(3) ACF and PACF of AR (p) model
For AR (p) model expression
Its ACF is
Wherein
Then
It is therefore seen that the ACF of AR (p) no matter k take it is much, to 1 to p rank lag ACF it is related, be not 0, for hangover sequence;
For PACF, calculated according to following:
For AR (1) process, as k=1,As k > 1,?
Exactly occurs peak value then truncation in k=1;
For AR (2) process, partial autocorrelation function is made of formula (9) and formula (10), in k≤2,Work as k > 2
When,The truncation namely after k=2;
Therefore, it releases for AR (p) process, in k≤p,As k > p, the PACF of AR (p) is 0, that is,
PACF is with the truncated sequence of p rank truncation;
(4) ACF and PACF of ARMA (p, q) model
The ACF of ARMA (p, q) model regards the mixing of the ACF of the ACF and AR (p) of MA (q) as;If p=0, it is truncated sequence,
If q=0, it is hangover sequence, if p and q are not 0, it is hangover sequence;
The PACF of ARMA (p, q) model regards the mixing of the PACF of the PACF and AR (p) of MA (q) as, if p=0, it is hangover sequence
Column, if q=0, it is truncated sequence, if p and q are not 0, it is hangover sequence;
S2-3, in the range that S2-2 step obtains p and q, different p and q values is combined, AIC criterion function is updated to
In calculated, the value of p and q when AIC minimum is obtained, so that it is determined that optimal ARIMA (p, d, q) model;AIC criterion function is such as
Under:
AIC=-2ln (L)+2k (21)
Wherein L is the maximum likelihood function value of ARIMA (p, d, q) model, and k is the estimative parameter of ARIMA (p, d, q) model
Number.
5. virtual machine bandwidth allocation methods of the OpenStack according to claim 1 based on application service, feature exist
In step S3 calculates the effective bandwidth based on application QoS according to application stream statistics characteristic, comprising the following steps:
S3-1, recalibration domain method calculate the Hurst index value H of flow sequence;
Specifically, recalibration domain method is also referred to as R/S analytic approach, the calculating of index value H is as follows:
(1) the time series X (t) that a length is N is given, dividing equally entire sequence with length n is A adjacent subintervals, then
There is A × n=N;Any subinterval is expressed as Ia, a=1,2 ..., A, in IaIn element representation be N (k, m), k=1,
2,......,n;M=1,2 ... .., A, IaMean value be
(2)IaFor the accumulation intercept X of mean valuek,aIt is defined as follows:
Ni,aIt indicates in IaIn element, i=1,2 ..., n
(3)IaIt is very poor
(4) subinterval IaStandard deviation
(5) eachBy correspondingIt is standardized, obtains sequence (R/S)n:
(6) n is since 3, and repeats step (1)-step (5) and obtain sequence [R/S] until n=Nn, n=3 ..., N;
(7) with Log (n) for explanatory variable, Log (R/S) is that explained variable carries out linear regression:
Log (R/S)=Loge+HLogn (26)
Wherein, e is the normal number randomly selected, and slope H is Hurst index value;
S3-2, the effective bandwidth flowed according to the statistical property of application stream, calculating application;
According to the statistical property of network flow, Y (t) is application traffic sequence, regards fractal Brown motion process as, then has:
Wherein, ZtIt is normal state fractal Brown motion, is also zero-mean, self-similarity nature parameter 0 < H < 1 and t2HFor the standard of variance
Fractal Brown motion, y are flow serial mean, and m is coefficient of variation, and t indicates time variable, and H is Hurst index value;
Derive that queue buffer queue length is greater than the probability of b by formula (27):
Wherein, C is effective bandwidth, k (H)=HH(1-H)1-H, P (X (t) > x) is equal to packet loss ε, and b indicates that queue buffer is big
Small, B (t) indicates that data reach the data length of buffer area with time change, substitutes into after formula (28) stream that is applied with ε
Effective bandwidth C:
Tolerant of delay and packet loss are considered in effective bandwidth calculating, overcome the originally effective bandwidth calculating side based on Loss Rate
Method cannot be guaranteed that the deficiency of time delay, binding cache time delay maximum value=queue buffer storage length ÷ queue output rate obtain formula
(30)
B=y × τ (30)
Wherein τ indicates the tolerant of delay of application stream, is substituted into formula (29) and obtains the effective bandwidth for meeting network QoS demand
Are as follows:
Wherein k=HH(1-H)1-H。
6. virtual machine bandwidth allocation methods of the OpenStack according to claim 1 based on application service, feature exist
It in, network load state described in step S4 is measured with network utilization μ, network utilization μ is virtual machine application stream
Effective bandwidth and administrator are the ratio of the bandwidth of virtual machine distribution;
Divide the network load state of virtual machine to low network load, high network load and full network load according to network utilization,
Wherein, if low network load indicates that following a period of time distributes current bandwidth, virtual machine network utilization rate μ will be less than
70%, need to recycle the bandwidth of 1- (μ+10%) at this time;If high network load indicates that following a period of time distributes current band
Width, virtual machine network utilization rate μ will be greater than or equal to 70% but be less than or equal to 85%, the bandwidth of application service virtual machine is not necessarily at this time
Change;If full network load indicates that following a period of time virtual machine distributes current bandwidth, virtual machine network utilization rate μ will be big
In 85%, the bandwidth of application service virtual machine needs to increase at this time, and initial increase is 1.1 times of original bandwidth allocation.
7. virtual machine bandwidth allocation methods of the OpenStack according to claim 1 based on application service, feature exist
In the method for salary distribution of step S5 is by current period bandwidth allocation E1, effective bandwidth C2According to formulaCalculate the following week
Phase virtual machine network utilization rate μ indicates the bandwidth that a virtual machine lower period should distribute with E, judges virtual machine according to μ value range
Future period network state, take β=0.7, γ=0.85
If 0 < μ < β, virtual machine carries out bandwidth recycling, recycles the bandwidth of 1- (μ+10%), then
E=E1× (μ+10%) (32)
If β≤μ≤γ, virtual machine bandwidth is remained unchanged, then
E=E1 (33)
If μ > γ, virtual machine bandwidth expands, then
E=E1× (1+10%) (34).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910487936.4A CN110365518A (en) | 2019-06-05 | 2019-06-05 | Virtual machine bandwidth allocation methods of the OpenStack based on application service |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910487936.4A CN110365518A (en) | 2019-06-05 | 2019-06-05 | Virtual machine bandwidth allocation methods of the OpenStack based on application service |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110365518A true CN110365518A (en) | 2019-10-22 |
Family
ID=68215635
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910487936.4A Pending CN110365518A (en) | 2019-06-05 | 2019-06-05 | Virtual machine bandwidth allocation methods of the OpenStack based on application service |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110365518A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110753009A (en) * | 2019-10-23 | 2020-02-04 | 重庆邮电大学 | Virtual machine and network bandwidth joint distribution method based on multi-QoS grouping |
WO2021098190A1 (en) * | 2019-11-20 | 2021-05-27 | 华为技术有限公司 | Time delay guarantee method, system and apparatus, and computing device and storage medium |
CN113890829A (en) * | 2021-11-19 | 2022-01-04 | 中徽建技术有限公司 | Network virtualization resource allocation method and system |
WO2022127456A1 (en) * | 2020-12-17 | 2022-06-23 | 中兴通讯股份有限公司 | Virtualized resource configuration method, apparatus and device, and storage medium |
CN115174405A (en) * | 2022-06-08 | 2022-10-11 | 西北大学 | Bandwidth allocation method based on ARIMA statistical model |
CN115866772A (en) * | 2023-02-16 | 2023-03-28 | 江西惜能照明有限公司 | Network bandwidth allocation method, device, medium and equipment based on intelligent lamp pole |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105515880A (en) * | 2016-01-25 | 2016-04-20 | 重庆邮电大学 | Token bucket traffic shaping method suitable for fusion network |
CN106161282A (en) * | 2015-03-26 | 2016-11-23 | 中国科学院上海高等研究院 | A kind of method of saving bandwidth resources hiring cost based on cloud computing environment |
CN106529145A (en) * | 2016-10-27 | 2017-03-22 | 浙江工业大学 | ARIMA-BP neutral network-based bridge monitoring data prediction method |
CN106844180A (en) * | 2017-02-07 | 2017-06-13 | 山东浪潮云服务信息科技有限公司 | A kind of monitoring and controlling forecast method of OpenStack platforms computing resource |
CN106920009A (en) * | 2017-03-03 | 2017-07-04 | 北京北青厚泽数据科技有限公司 | The Forecasting Methodology of hospital services amount |
CN107633331A (en) * | 2017-09-26 | 2018-01-26 | 北京福布罗科技有限公司 | Time series models method for building up and device |
-
2019
- 2019-06-05 CN CN201910487936.4A patent/CN110365518A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106161282A (en) * | 2015-03-26 | 2016-11-23 | 中国科学院上海高等研究院 | A kind of method of saving bandwidth resources hiring cost based on cloud computing environment |
CN105515880A (en) * | 2016-01-25 | 2016-04-20 | 重庆邮电大学 | Token bucket traffic shaping method suitable for fusion network |
CN106529145A (en) * | 2016-10-27 | 2017-03-22 | 浙江工业大学 | ARIMA-BP neutral network-based bridge monitoring data prediction method |
CN106844180A (en) * | 2017-02-07 | 2017-06-13 | 山东浪潮云服务信息科技有限公司 | A kind of monitoring and controlling forecast method of OpenStack platforms computing resource |
CN106920009A (en) * | 2017-03-03 | 2017-07-04 | 北京北青厚泽数据科技有限公司 | The Forecasting Methodology of hospital services amount |
CN107633331A (en) * | 2017-09-26 | 2018-01-26 | 北京福布罗科技有限公司 | Time series models method for building up and device |
Non-Patent Citations (2)
Title |
---|
丁培培: "汇率波动的复杂性研究", 《中国优秀硕士学位论文全文数据库社会科学Ⅰ辑》 * |
籍风磊: "面向业务的带宽估计和带宽分配方法的研究", 《中国优秀硕士毕业论文全文数据库信息科技辑》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110753009A (en) * | 2019-10-23 | 2020-02-04 | 重庆邮电大学 | Virtual machine and network bandwidth joint distribution method based on multi-QoS grouping |
CN110753009B (en) * | 2019-10-23 | 2022-08-26 | 重庆邮电大学 | Virtual machine and network bandwidth joint distribution method based on multi-QoS grouping |
WO2021098190A1 (en) * | 2019-11-20 | 2021-05-27 | 华为技术有限公司 | Time delay guarantee method, system and apparatus, and computing device and storage medium |
WO2022127456A1 (en) * | 2020-12-17 | 2022-06-23 | 中兴通讯股份有限公司 | Virtualized resource configuration method, apparatus and device, and storage medium |
CN113890829A (en) * | 2021-11-19 | 2022-01-04 | 中徽建技术有限公司 | Network virtualization resource allocation method and system |
CN115174405A (en) * | 2022-06-08 | 2022-10-11 | 西北大学 | Bandwidth allocation method based on ARIMA statistical model |
CN115866772A (en) * | 2023-02-16 | 2023-03-28 | 江西惜能照明有限公司 | Network bandwidth allocation method, device, medium and equipment based on intelligent lamp pole |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110365518A (en) | Virtual machine bandwidth allocation methods of the OpenStack based on application service | |
CN107566194B (en) | Method for realizing cross-domain virtual network mapping | |
JP6162046B2 (en) | A system that shares network bandwidth among competing applications | |
Kumar et al. | BwE: Flexible, hierarchical bandwidth allocation for WAN distributed computing | |
CN109995669B (en) | Distributed current limiting method, device, equipment and readable storage medium | |
CN107124445A (en) | A kind of collecting method and device | |
CN108809857A (en) | A method of the traffic monitoring based on SDN and service quality securing strategy | |
CN109964507B (en) | Network function management method, management unit and system | |
WO2014090075A1 (en) | System and method for estimating an effective bandwidth | |
CN107018024B (en) | Cloud service recommendation method and device | |
CN103309723B (en) | Virtual machine resource integration and method | |
HK1126052A1 (en) | Aggregated resource reservation for data flows | |
CN112217725B (en) | Delay optimization method based on edge calculation | |
CN105897612A (en) | Multi-service dynamic bandwidth allocation method and system based on SDN (Software Defined Network) | |
CN105992233A (en) | Method and device for calculating capacity baseline of network service | |
WO2016095410A1 (en) | Link traffic distributing method and device | |
CN103384251A (en) | Multi-class safety service combinable safety network construction method and device | |
Sharma et al. | Evaluating performance attributes of layered software architecture | |
CN104363187A (en) | Gateway resource response method and gateway resource response device for internet of things | |
Gelenbe et al. | Cognitive users with useful vacations | |
Souza et al. | Towards a Fog-to-Cloud control topology for QoS-aware end-to-end communication | |
Riska et al. | Efficient fitting of long-tailed data sets into phase-type distributions | |
Stüdli et al. | On the AIMD algorithm under saturation constraints | |
Jiang et al. | Analysis of stochastic service guarantees in communication networks: A traffic model | |
JP2004302525A (en) | Device and method for carrying out performance balance evaluation and sizing of component of computer system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191022 |